The Agent Skills Directory

[COMMAND_EXECUTION]: The tool uses subprocess.run to interact with git for repository metadata extraction. This execution is strictly scoped to the git binary and is required for the tool's core functionality of repository orientation.- [DATA_EXFILTRATION]: The skill reads local files to build architectural insights. All findings are written to user-specified local directories. There is no evidence of unauthorized data access, credential harvesting, or exfiltration to remote servers.- [INDIRECT_PROMPT_INJECTION]: As an analysis tool, the skill processes untrusted source code as data. It mitigates injection risks by using structured data formats (JSONL) and providing explicit boundary labels for evidence versus inference, allowing the agent to parse the output without conflating source code with instructions.- [SAFE]: No malicious patterns such as obfuscation, persistence, or privilege escalation were detected. The skill performs its stated function using standard Python libraries and follows security best practices like output sanitization for its generated views.

zj-code-research